Belgium e-Procurement Tenders Scraper: BOSA Notices

Monitor Belgian public procurement: Benelux and EU bid teams, consultants and suppliers track new overheidsopdrachten / marchés publics on BOSA e-Procurement (publicprocurement.be) by keyword, CPV, NUTS region, buyer, notice type or date, as clean rows.

Live sample, pulled 2026-11-06 (noticeTypes: ["16","17"], publishedFrom: 2026-11-01), 12 rows in sample.json:

noticeId	title	buyer	KBO/BCE no.	mainCpv	published	noticeUrl
46fa7ef2-35cd-4749-8bf5-947022aecc77	Cahier Spécial des Charges n°57/53/4/26/083 - Ligne 43 : Riv	Infrabel Projects South	0869763267	45220000	2026-11-06	https://www.publicprocurement.be/publication-workspaces/46fa7ef2-35cd-4749-8bf5-947022aecc77/general
62361823-708a-40ff-93ba-519ab050b2cf	Zuidijzerpolder: Onderhoudswerken aan onbevaarbare waterlope	ZUIDIJZERPOLDER	0851235970	45240000	2026-11-06	https://www.publicprocurement.be/publication-workspaces/62361823-708a-40ff-93ba-519ab050b2cf/general
7291a5d2-15ff-48ee-a598-4b73b8ccca8b	57/53/4/26/145 L166 : Gedinne -Bertrix Commune de Bertrix En	Infrabel Projects South	0869763267	45000000	2026-11-06	https://www.publicprocurement.be/publication-workspaces/7291a5d2-15ff-48ee-a598-4b73b8ccca8b/general
2bc73dfb-4ddd-4b39-a50c-3a8bbbe1d35c	Achat d'un camion d'occasion équipé d'une plate-forme élévat	Stadt Sankt Vith	0207078568	34137000	2026-11-05	https://www.publicprocurement.be/publication-workspaces/2bc73dfb-4ddd-4b39-a50c-3a8bbbe1d35c/general
b1859abc-4fc0-4904-aa8a-f8ba389be4fb	Slibruiming van de Aa te Vorselaar en Lille	Vlaamse Milieumaatschappij	0887290276	45246000	2026-11-05	https://www.publicprocurement.be/publication-workspaces/b1859abc-4fc0-4904-aa8a-f8ba389be4fb/general

Text is kept as served (NL/FR/DE/EN). title is the first available in the order EN > NL > FR > DE, language says which; titles has every language version. Empty values (null) are not published by the buyer.

Input

field	example	meaning
keywords	"informatique"	free-text search
cpv	["72000000"]	list of 8-digit CPV codes
noticeTypes	["16","17"]	BOSA notice subtype codes (16/17 contract notices, 20 modification, E1-E6 eForms); empty = all
location	"BE10"	Belgian NUTS code (BE10 Brussels, BE2 Flanders, BE3 Wallonia, BE332 Liège)
buyer	"Infrabel"	buyer name contains (filtered client-side)
publishedFrom / publishedTo	"2026-10-01"	publication date range, YYYY-MM-DD
deadlineFrom	"2026-11-10"	tender deadline on or after (server-side; notices without deadline drop out)
includeDetails	true	default true; one extra request per notice for KBO/BCE number, value, documents link
maxItems	50	default 50, max 1000

Newest first.

Output

field	meaning
noticeId	BOSA publication workspace id (UUID, unique)
referenceNumber	buyer's file reference
title, titles, language, languages	title (EN>NL>FR>DE), all title versions {lang: text}, chosen language, publication languages
description	as served; e-mail addresses stripped
buyerName, buyerOrgNumber	buyer and Belgian KBO/BCE company number (10 digits; needs includeDetails)
noticeType, procedureType, nature, status	e.g. 16, OPEN, WORKS/SUPPLIES/SERVICES, ACTIVE
cpvCodes (array), mainCpv	8-digit CPV
nutsCodes	NUTS codes
estimatedValue, currency	only when the notice states a value (needs includeDetails); mostly null
publicationDate, deadline	date / local timestamp
noticeUrl, documentsUrl	publicprocurement.be page, tender documents link (needs includeDetails)
scrapedAt	ISO timestamp

Pricing

Pay per event: $0.005 per start (actor-start) + $0.002 per notice (notice). Example: 1000 notices = $0.005 + 1000 x $0.002 = $2.005. The run stops cleanly at your spending cap; nothing is pushed beyond maxItems.

Source and limits

Step 0 findings (2026-11-06): publicprocurement.be is a Vue SPA calling its own JSON APIs (no registration, no captcha). Plain calls without a token get 403; the SPA first takes an anonymous visitor token, which the actor does the same way:

• GET https://www.publicprocurement.be/env.config.js publishes the public frontend client id/secret.
• POST https://www.publicprocurement.be/auth/realms/supplier/protocol/openid-connect/token (grant_type=client_credentials, Basic auth with that public client) -> 200, 1 h visitor token. No account or private key involved.
• POST https://www.publicprocurement.be/api/sea/search/publications -> 200. Body {terms, cpvCodes, nutsCodes, noticeSubtypes, publicationDateFrom/To, tenderSubmissionDeadlineFrom/To, page, pageSize<=100}, needs Authorization: Bearer + BelGov-Trace-Id.
• GET https://www.publicprocurement.be/api/dos/publication-workspaces/{noticeId} -> 200 (eForms XML: KBO/BCE number, documents URL, deadline).

Limits: unofficial API that can change; the visitor-token route relies on the site's public client, so it may be closed without notice; buyer filter applies after the search; the source rarely states estimated values; max 1000 items per run. No proxy.

Business data only

Buyer organisation, notice, CPV, value, dates and URLs. Contact persons, personal e-mails and phones are not read or output.

Run / test

python3 check.py (offline against fixtures/belgium.json, real BOSA responses; asserts query mapping, maxItems, PPE charge counts, no personal data). python3 check.py --live makes 1 real search request, needs 5+ notices (last result: live.log, 2026-11-06). Run: put input in storage/key_value_stores/default/INPUT.json, node src/main.js.

Publishing

publish.py --noruns (needs APIFY_TOKEN) creates/updates the actor without test runs. Not run: #469's publish.py was not run either, so this one is ready, not published.